# CompressKV: Semantic Retrieval Heads Know What Tokens are Not Important Before Generation

arXiv: https://arxiv.org/abs/2508.02401 · HTML: https://mlsanity.com/paper/2508.02401/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Placed in KV cache optimization, one of 12 directions in Efficient LLM inference and serving.
- Builds on 15 of its 23 references that are in the map.
- First posted on arXiv 4 Aug 2025.

## Placed in

- Efficient LLM inference and serving / [KV cache optimization](https://mlsanity.com/inference/kv-cache-optimization/)

## Builds on (most cited first, 15 of 15)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning](https://mlsanity.com/paper/2307.08691/) (2023)
- [Efficient Streaming Language Models with Attention Sinks](https://mlsanity.com/paper/2309.17453/) (2023)
- [LongBench: A Bilingual, Multitask Benchmark for Long Context Understanding](https://mlsanity.com/paper/2308.14508/) (2023)
- [In-context Learning and Induction Heads](https://mlsanity.com/paper/2209.11895/) (2022)
- [SnapKV: LLM Knows What You are Looking for Before Generation](https://mlsanity.com/paper/2404.14469/) (2024)
- [KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache](https://mlsanity.com/paper/2402.02750/) (2024)
- [Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs](https://mlsanity.com/paper/2310.01801/) (2023)
- [Scissorhands: Exploiting the Persistence of Importance Hypothesis for LLM KV Cache Compression at Test Time](https://mlsanity.com/paper/2305.17118/) (2023)
- [DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads](https://mlsanity.com/paper/2410.10819/) (2024)
- [A Fast Post-Training Pruning Framework for Transformers](https://mlsanity.com/paper/2204.09656/) (2022)
- [PyramidInfer: Pyramid KV Cache Compression for High-throughput LLM Inference](https://mlsanity.com/paper/2405.12532/) (2024)
- [CAKE: Cascading and Adaptive KV Cache Eviction with Layer Preferences](https://mlsanity.com/paper/2503.12491/) (2025)
- [Basis Sharing: Cross-Layer Parameter Sharing for Large Language Model Compression](https://mlsanity.com/paper/2410.03765/) (2024)

## Built on by (most cited first, 9 of 9)

- [KV Packet: Recomputation-Free Context-Independent KV Caching for LLMs](https://mlsanity.com/paper/2604.13226/) (2026)
- [RedKnot: Efficient Long-Context LLM Serving with Head-Aware KV Reuse and SegPagedAttention](https://mlsanity.com/paper/2606.06256/) (2026)
- [TPLA: Tensor Parallel Latent Attention for Efficient Disaggregated Prefill and Decode Inference](https://mlsanity.com/paper/2508.15881/) (2025)
- [Adaptive Filtering of the KV Cache: Diagnosing and Correcting Structural-Role Bias in LLM Inference](https://mlsanity.com/paper/2607.13205/) (2026)
- [RippleKV: Cross-Layer KV Cache Allocation via Perturbation Propagation](https://mlsanity.com/paper/2608.08684/) (2026)
- [Trust the Mass: Forced Weights in KV-Cache Eviction](https://mlsanity.com/paper/2608.25230/) (2026)
- [Dustin: Draft-Augmented Sparse Verification for Efficient Long-Context Generation with Speculative Decoding](https://mlsanity.com/paper/2606.24957/) (2026)
- [Logit-Contribution Scoring Identifies Non-Literal Retrieval Heads](https://mlsanity.com/paper/2607.01002/) (2026)
- [Training-Free Hashing-Based Attention via Binary Principal Components](https://mlsanity.com/paper/2608.04405/) (2026)

## Abstract (arXiv)

Recent advances in large language models (LLMs) have significantly boosted long-context processing. However, the increasing key-value (KV) cache size poses critical challenges to memory and execution efficiency. Most KV cache compression methods rely on heuristic token eviction using all attention heads in Grouped Query Attention (GQA)-based LLMs. This method ignores the different functionalities of attention heads, leading to the eviction of critical tokens and thus degrades the performance of LLMs. To address the issue above, instead of using all the attention heads in GQA-based LLMs to determine important tokens as in the previous work, we first identify the attention heads in each layer that are not only capable of retrieving the initial and final tokens of a prompt, but also capable of retrieving important tokens within the text and attending to their surrounding semantic context. Afterwards, we exploit such heads to determine the important tokens and retain their corresponding KV cache pairs. Furthermore, we analyze the cache eviction error of each layer individually and introduce a layer-adaptive KV cache allocation strategy. Experimental results demonstrate the proposed CompressKV consistently outperforms state-of-the-art approaches under various memory budgets on LongBench and Needle-in-a-Haystack benchmarks. Our code is publicly available at: https://github.com/TUDa-HWAI/CompressKV.git.